A forecast is a plan-ready estimate, not a promise. VisitorCast shows you four things at once: how many visitors to expect, how uncertain that expectation is, what is driving it, and how well past forecasts have held up. Read together, they tell you how much weight to put on the numbers.

Choose your forecast length first

At the top of the Forecast page you choose between 7 days, 30 days and 90 days. The choice is not only about how far ahead you look — it also changes how precise the forecast can be.

Length Best for What to expect
7 days Staffing and rostering, front-of-house cover, guided tours, café and shop shifts The most accurate view. Exhibitions, events and holidays in the coming week are all known.
30 days Operations, monthly budgets, stock ordering, volunteer scheduling The everyday planning horizon, and the default when you open the page.
90 days Marketing and campaign timing, seasonal staffing, board and budget conversations Direction and seasonal shape rather than day-level precision. Use it to spot the quiet weeks worth promoting.

A forecast is saved per length, so switching between the tabs shows the last forecast generated for that length. Select Generate to run a fresh one on your latest data.

The trend card: the one-line answer

Above the chart, a single sentence compares the whole period with the same period last year — for example, "Next 30 days are expected to bring 1,250 more visitors than the same period last year."

Green means up, red means down, grey means roughly flat. This is the number to bring to a meeting. Everything below it explains where it comes from.

The chart: the line and the band

The chart shows two things.

  • The predicted line — the expected number of visitors for each day. This is the number to plan from.
  • The confidence band — the shaded area around the line, showing the range the actual number is expected to fall within on 8 out of 10 days.

The band is always set to 80%, so it means the same thing on every forecast and every museum. Use it as a rough sense of how much room to leave around the predicted number, not as a separate signal to interpret.

A dashed line shows last year's actual visitors for the same days. It is context, not part of the calculation. If the forecast sits far from last year, the driver cards below should explain why.

How "last year" is matched

Comparing a Saturday with a Saturday matters more than comparing 12 July with 12 July. A museum's week has a strong rhythm, and a calendar date that lands on a Tuesday this year may have been a Sunday last year.

So VisitorCast does not compare against the same calendar date. It compares against the matching weekday 52 weeks earlier — 364 days back rather than 365. A Saturday is always compared with a Saturday, and school vacations, holiday weekends and busy periods line up far more closely than they would on a date-for-date comparison.

A day only appears in the comparison if it is genuinely comparable. If the museum was closed on that day 52 weeks ago, or there is no usable visitor number on record for it, VisitorCast leaves the cell blank rather than comparing against a day that was not equivalent. That is why the Last year column has gaps, and why the trend card compares only the days it could actually match.

The table: day by day

Under the chart, Forecast by day lists every day in the period. The header tells you how many days are open and how many are closed.

  • Date — with the weekday in brackets. Weekend dates are highlighted.
  • Predicted — the expected visitor count for that day.
  • Range — the low and high end of the confidence band for that day.
  • Last year — the actual count on the matching weekday 52 weeks earlier. Blank if that day is not comparable.
  • vs last year — the percentage difference, coloured green or red when the gap is meaningful.
  • Day type — labels such as the name of a public holiday, Weekend, or Closed.

Closed days show a dash instead of a number. VisitorCast never predicts visitors for a day you are not open, and closed days are left out of the totals.

This table is the one to export or screenshot when you are building a roster or a shift plan. The Range column is what turns a forecast into a staffing decision — see how to use a visitor forecast for staffing and planning.

The driver cards: why the forecast looks like this

What's driving this forecast shows the factors with the largest effect on the period total, ranked by impact. Each card names the driver, says whether it boosts or reduces the forecast and by how many visitors, and gives a plain-language reason — for example, that school vacations fall in this period, or that the exhibitions running are less popular than a typical week for your museum.

The number of cards varies. Only drivers that are actually active in the period appear, so a quiet month with no holidays, vacations or exhibitions shows fewer cards than a busy one. You are seeing the drivers that matter for this forecast, not a fixed list.

These numbers come straight from the model's own arithmetic. Because the model is additive, every driver contributes a concrete number of visitors, and those contributions add up to the forecast. See how VisitorCast predicts visitor numbers for how the layers fit together.

Which drivers are available depends on what you have enabled on the Drivers page. If a factor you care about is missing from the cards, check that its signal is turned on and that your exhibitions and events are up to date.

The track record: how much to trust it

At the bottom of the page, Forecast track record compares every past forecast with the visitor numbers you later confirmed. It shows three figures for the selected forecast length:

  • Days close to the mark — how many confirmed days landed within 10% of the actual count.
  • A single day, on average — the average miss on an individual day.
  • The whole period, on average — the same comparison on the period total. This figure is usually better, because days that land a little high and a little low cancel each other out.

That difference matters for how you use the forecast. Period totals are reliable earlier than individual days, so a monthly budget built from a 30-day forecast is on firmer ground than a single Tuesday's roster.

VisitorCast needs around 25 confirmed days per forecast length before it reports a track record. Until then the card shows how far along it is. If you want an answer sooner, run an accuracy check — it predicts a period that has already happened and shows the forecast next to your real numbers.

As a rough guide, an average miss of about 15% on a 7-day forecast and about 20% on a 30 or 90-day forecast is a good result. Planning from gut feeling alone is often 30–40% off.

A forecast is a guess, not a fact

Before you plan on any of these numbers, hold on to one thing: a forecast is the best estimate VisitorCast can make from your data — never the truth.

It knows your visitor history, your opening days, your exhibitions and events, and the signals you have enabled. It knows nothing about a rail strike, a burst pipe, a heatwave, a cancelled coach party or a festival across town. Your team knows about those. The model never will.

So read the forecast as the most informed opinion available, not as an instruction. Where you know something it cannot, you are right and it is wrong. And when you pass a forecast on to others, present it as a projection with a range — a number presented as certain will be remembered as a promise.

How to use a visitor forecast for staffing and planning covers how to build that uncertainty into real decisions.

A quick reading routine

  1. Read the trend card — is the period up or down against last year?
  2. Scan the chart for the shape: which weeks are busy, which are quiet?
  3. Check the driver cards — does the explanation match what you know is happening?
  4. Open the table for the days you are actually planning, and note the range.
  5. Glance at the track record to decide how much buffer to build in.

If step 3 does not match reality — a big exhibition is missing, an event is not recorded — fix the underlying data and regenerate. The forecast is only as good as what it knows about. See how to get the best forecasts.